Differential Geometry guide - •• Luis Floritluis.impa.br/aulas/geodif/aulas.pdf · Luis A....
Transcript of Differential Geometry guide - •• Luis Floritluis.impa.br/aulas/geodif/aulas.pdf · Luis A....
Differential Geometry guide
Luis A. Florit ([email protected], office 404)
PART IPrerequisites (part I): Analysis on R
n.Analysis on manifolds, Stokes and de Rham recommended. Bibliography: [dC]
§1. Introduction
Differentiable manifolds: smooth world. Now we’re going to
measure in them. After all, geometry comes from the Greek:
“measurement of the Earth”:
Eratosthenes (Cirene, 276 AC - Alexandria, 194 AC)
Posidonius (135 AC - 51 AC) ⇒ Colombo
We will study in this first part curves, surfaces and hypersurfaces
of Euclidean space ⇒ Two aspects: intrinsic and extrinsic.
§2. Curves
Curve: intrinsically, nothing interesting: I ⊆ R.
Regular curves α in R2 and R
3: arc length s, p.b.a.l, curvature
κα and torsion τα. Frenet Trihedron: t, n, b.FTC: explicit vs. ODE.
Curves in Rn: FTC.
Exercise. Prove that κα =√
‖α′‖2‖α′′‖2 − 〈α′, α′′〉2/‖α′‖3, indepen-
dently of the parametrization of α.
1
§3. Surfaces in R3
Regular (Euclidean) surface: S = S2 ⊂ R3 (embedding!)
Regular (Euclidean) hypersurface: Mn ⊂ Rn+1 (embedding!)
Regular (Euclidean) submanifold: Mn ⊂ Rn+p (embedding!)
It is enough to check that: ∀x ∈ Mn, ∃V ⊂ Rn+p open, x ∈ V ,
and a smooth map U ⊂ Rn 7→ Mn∩V ⊂ R
n+p that is injective,
open and has rank n: coordinates (smooth = Cr/C∞/Cw).
Examples:
graf(f ) for f : U ⊂ Rn → R.
g−1(t0) for a regular value t0 (in the image) of g : W ⊂ Rn+1 → R:
Sphere Sn ⊂ R
n+1.
Ellipsoid g−1(r) for g(x, y, z) = x2/a2 + y2/b2 + z2/c2, r > 0.
Hyperboloid g−1(r) for g = x2 + y2 − z2 (Hyperboloid of two
sheets for r < 0, while the Cone g−1(0) is NOT a regular surface).
Hyperbolic paraboloid g−1(0) for g = x2/a2 − y2/b2 − z.
Circular cylinder and Cylinders over conics.
Def.: Parametrized surface just ϕ : U ⊂ R2 → R
3. Singular
points of ϕ: dϕp singular. We call ϕ regular if it is an immersion.
Proposition 1. Every hypersurface is locally a graph.
Exercise. No need to check that a coordinate system is open (homeomor-
phism) if we know beforehand that Mn ⊂ Rn+1 is a regular hypersurface.
Differentiable functions: now we can use the ambient space.
2
Examples: F : U⊂R3→R smooth ⇒ F |S is smooth ∀S ⊂ U
ϕ : U ⊂ R2 → R
3 coordinates ⇒ U and ϕ(U) are diffeomorphic
S symmetric ⇒ the symmetry restricted to S is smooth:
Example: Surfaces of revolution: Meridians, Parallels, Axis,
Generatrix (embedded!).
Example: Tangent surfaces to a curve α : I → R3: κα 6= 0
⇒ ϕ(s, t) = α(s) + tα(s) is a parametrized surface, regular for
t 6= 0.
§4. Tangent space as a subspace
For a regular submanifoldMn ⊂ Rn+q and p ∈ Mn, we now have
TpM naturally included in Rn+q as an affine subspace: spanned
by (∂ϕ/∂ui)(p) : i = 1, . . . , n for any coordinate ϕ at p.
If α :I→S ⊂ R3, α(s) = ϕ(u(s), v(s)), α(0) = p = ϕ(0, 0) ⇒
α′(0) = u′(0)∂ϕ
∂u(0, 0) + v′(0)
∂ϕ
∂v(0, 0) ∈ TpS ⊂ R
3
Differential of a function f : S1 ⊂ R3 → S2 ⊂ R
3 at a point: can
be seen as a linear map between subspaces of R3.
Local diffeomorphism, chain rule...
TIP: Use curves to compute differentials!
Example: L : R3 → R3 linear, S ⊂ R
3 ⇒ f∗p = f |TpSSince we fixed an orientation in R3, we can talk about the Normal
3
vector field of our surface:
N =∂ϕ∂u ∧
∂ϕ∂v
‖∂ϕ∂u ∧
∂ϕ∂v‖
(1)
Same holds for hypersurfaces.
Angle between surfaces.
§5. The First Fundamental Form
Curves + inner product 〈 · , · 〉 of R3 → distance → first funda-
mental form I :
I(p) = 〈 · , · 〉|TpS×TpS : TpS × TpS → R
is an inner product on TpS.
We will denote also by I its associated quadratic form.
Arc length s for α : I → S ⊂ R3:
s(t) =
∫ t
t0
√
I(α′(r))dr.
ϕ : U → S ⊂ R3 coordinate system ⇒
E = ‖∂ϕ∂u
‖2, F = 〈∂ϕ∂u
,∂ϕ
∂v〉, G = ‖∂ϕ
∂v‖2 ∈ C∞(U)
coefficients of I in the coordinate ϕ.
Im (α) ⊂ Im (ϕ) ⇒ s(t) =∫ t
t0
√Eu′2 + 2Fu′v′ +Gv′2dr.
Remark 2. I is a symmetric positive definite (2,0)-tensor in S,
a Riemannian metric on S: I = inc∗〈 · , · 〉.
4
Same for arbitrary regular Euclidean submanifolds:
Example: Affine plane through p ∈ R3 or Cylinder over plane
curve ⇒ everywhere coordinate systems with E≡1≡G, F ≡0.
Def.: Regular domain D ⊂ S.
If ϕ : U → S is a coordinate system, ‖∂ϕ∂u ∧
∂ϕ∂v‖ is the area of the
parallelogram determined by the coordinate vector fields, and we
can define the area of a regular domain Ω ⊂ ϕ(U) by
A(Ω) =
∫
ϕ−1(Ω)
‖∂ϕ∂u
∧ ∂ϕ
∂v‖ =
∫
ϕ−1(Ω)
√
EG− F 2 dudv
since ‖x ∧ y‖2 + 〈x, y〉2 = ‖x‖2‖y‖2
§6. Recalling basic concepts of vector bundles
§7. Orientation
graf(f ) orientable
Mn ⊂ Rn+1 orientable ⇐⇒ there exist a globally defined smooth
unit normal vector field.
g−1(r) orientable (0 6= grad (g) ⊥ M)
Theorem 3. Mn ⊂ Rn+1 embedded, orientable ⇒ there
exist V ⊂ Rn+1 open with Mn ⊂ V , and g : V ⊂ R such that
0 is a regular value of g and Mn = g−1(0).
Proof: Existence of tubular neighborhoods.
Remark 4. Mn ⊂ Rn+1 embedded and compact ⇒ orientable
(Jordan Theorem 41 in our last course).
5
§8. Gauss map and Second Fundamental Form
For any Euclidean hypersurface Mn ⊂ Rn+1, locally we have a
unit normal vector field
N : U ⊂ Mn → Sn ⊂ R
n+1.
But TpM is parallel to TN(p)Sn, and hence dNp ∈ End(TpM).
Moreover, dNp is self adjoint (w.r.t. I), so the quadratic form
IIp(w) := −〈dNpw,w〉
is called the second fundamental form of Mn at p. We also give
the same name to the associated symmetric tensor,
Ap := −dNp.
Def.: Let α : I → M be a regular curve through p = α(0) ∈ M .
The normal curvature of α at p is given by
κn := κ〈N, n〉,
where n is the normal vector of α at p and κ its curvature.
Remark 5 . (!!!!) It holds that (draw a picture)
II(α′(0)) = κn(0)
i.e., the normal curvature only depends on the direction of α′ (!!).In particular, if v ∈ TpM and π is the plane spanned by v and
N(p), α = π ∩M ⊂ π is a plane curve whose curvature is II(v)
(beware of orientations).
6
Examples: Second fundamental form of a graph; y = x4; Sn.
A self adjoint ⇒ principal curvatures ki, principal directionsei, lines of curvature:
v =∑
i
viei ⇒ II(v) =∑
i
kiv2i .
For dimension 2, we have the Euler’s formula for ‖v‖ = 1:
v = cos(θ)e1 + sin(θ)e2 ⇒ II(v) = k1 cos(θ)2 + k2 sin(θ)
2.
Remark 6. Ordering the principal curvatures k1 ≤ · · · ≤ knwe see by Remark 5 that e1 is the direction where M “curves”
less (w.r.t. N) in the ambient space, while en is the one where it
curves more. This follows from the usual diagonalization process.
§9. The two curvatures for surfaces: K and H
For a symmetric endomorphism in dimension two, we have two
invariants (independent of orthonormal basis): the trace and the
determinant.
Def.: For a regular surface S ⊂ R3 and p ∈ S, the Gaussian
curvature of S at p is given by
K(p) := det(Ap) = k1k2.
Def.: The mean curvature of S at p is H(p) = −trace(Ap)/2.
We will see in a while that the two curvatures have very different
nature.
7
Notice that K does not depend on orientation, while H does.
Notice that both are smooth functions that determine k1 and k2:
ki = H ±√
H2 −K.
Def.: A point p in S can be elliptic, hyperbolic, parabolic,
planar (or totally geodesic), minimal, umbilical. Accordingly,
S itself could be totally geodesic, minimal, umbilical.
Remark 7. The principal curvatures and their eigenspaces are
always continuous, and smooth along any open subset where their
multiplicities are constant. In particular, they are always smooth
along (the connected components of) an open dense subset W of
M . For surfaces, if V is the set of umbilical points of S, W can
be taken as V o ∪ S \ V .
Def.: Asymptotic direction and asymptotic curve of S ⊂ R3.
Notice that there exists an asymptotic direction at p ⇐⇒K(p) ≤ 0, while there are precisely two asymptotic directions
at p ⇐⇒ K(p) < 0.
Proposition 8. Let Mn ⊂ Rn+1 regular and connected.
Then, M is umbilical ⇐⇒ Mn is (an open subset of) a
round n-sphere or a hyperplane.
Def.: Conjugate directions.
8
§10. II and K in coordinates, part 1
For a coordinate system ϕ = ϕ(u, v) in a surface S ⊂ R3, let as
before
N =∂ϕ∂u ∧
∂ϕ∂v
‖∂ϕ∂u ∧
∂ϕ∂v‖
.
Denote by (aij) the matrix of A in the coordinate basis,
Nu = a11ϕu + a12ϕv,
Nv = a21ϕu + a22ϕv.
Define the functions
e := −〈Nu, ϕu〉 = 〈N,ϕuu〉,
g := −〈Nv, ϕv〉 = 〈N,ϕvv〉,f := −〈Nu, ϕv〉 = −〈Nv, ϕu〉 = 〈N,ϕuv〉.
Hence, if v = v1ϕu + v2ϕv, II(v) = ev21 + 2fv2v2 + gv22, and(
a11 a12a21 a22
)
=−1
EG− F 2
(
e f
f g
)(
G −F
−F E
)
These are known as the Weingarten equations. In particular,
for the Gaussian curvature we obtain
K =eg − f 2
EG− F 2.
Example: The torus. For 0 < r < a, take the (almost global)
chart
ϕ(u, v) = ((a + r cos(u)) cos(v), (a + r cos(u)) sin(v), r sin(u)) .
9
Hence,N = (cos(u) cos(v), cos(u) sin(v), sin(u)) (geometrically!),
and hence E = r2, G = (a + r cos(u))2, F = 0, and
K =cos(u)
r(a + r cos(u)).
Make the computation, make a picture, sign K interpretation,
elliptic/hyperbolic points, move a and r and see how K varies,
independence of v... see everything geometrically!
Remark 9. Observe that∫
K = 0, independently of a and r!
Now, let’s compute K for any surface of revolution of a simple
closed curve p.b.a.l. as (a(s), 0, b(s)): K = −a′′/a,∫
K = 0!!!!
Proposition 10. If p ∈ S ⊂ R3 is elliptic ⇒ a neighborhood
of p is in one side of TpS. If p hyperbolic, it is not.
Proof: Differentiate g = 〈ϕ− ϕ(0), N(p)〉 at p = ϕ(0).
Def.: Lines of curvature.
If α(t) = ϕ(u(t), v(t)) is a line of curvature, then dN(α′)) = λα′,or equivalently, (fF − eG)u′ + (gF − fG)v′ = λu′(EG − F 2),
(eF − fE)u′ + (fF − gE)v′ = λv′(EG− F 2), or
(fE − eF )(u′)2 + (gE − eG)u′v′ + (gF − fG)(v′)2 = 0,
known as the equation of the lines of curvature. In particular,
outside of the umbilical points:
the chart is by lines of curvature ⇔ F = f = 0.
10
Proposition 11. Let p ∈ S ⊂ R3, and a sequence of compact
regions Bi ⊂ S such that Bi → p. Then,
|K(p)| = limi→∞
Area(N(Bi))
Area(Bi).
Proof: Follows from |Nu∧Nv| = |K||ϕu∧ϕv|, even ifK(p) = 0.
Remark 12. At the non-flat points, N preserves orientation if
and only ifK > 0. Hence we can remove the modulus if we define
“oriented area”.
§11. Vector fields
Recall: Trajectories (= integral curves), F.T. ODE, local flux.
X(M) = Γ(TM)
For f : N → M , Xf = Γ(f ∗(TM)).
Exercise. If X, Y ∈ X(M) and ξ is the local flux of X around p, then
[X, Y ](p) = limt→01t((ξ−t)∗Y (ξt(p))− Y (p)) .
Lemma 13. Let M be any manifold, p ∈ M and X ∈ X(M)
with X(p) 6= 0. Then, there is a coordinate system around
p such that X|U = ∂/∂x1. In particular, if M is a surface,
there is a first integral of X in U , i.e., a function f : U → R
with dfp 6= 0 such that f is constant along trajectories of X.
Proof: Use the flux of X to construct a suitable chart in p for
which X|U is a coordinate vector field.
Lemma 14. Same hypothesis as in Lemma 13, and g ∈F(U) ⇒ there is µ ∈ F(U), µ > 0, such that X(µ) = g.
11
Proof: Use the flux of X to write this as an ODE.
Lemma 15. Two vector fields X, Y satisfy that [X, Y ] = 0
if and only if their fluxes commute: φXt φY
s = φYs φX
t ∀t, s.Proof: At points whereX = Y = 0 it is obvious, so assumeX 6=0. By the ‘ϕ-related’ property for Lie brackets of vector fields, we
can assume our manifold is Rn. Moreover, by Lemma 13, we can
assume X = ∂/∂x1 ∼= e1, and the lemma follows easily.
Lemma 16. Same hypothesis as in Lemma 13 and Y ∈ X(S)
linearly independent with X in p ⇒ there is a chart ϕ around
p whose coordinate vector fields are colinear with X and Y .
Proof: Write [X, Y ] = gX + fY and use Lemma 14 to find
µ, λ positive functions such that [X ′, Y ′] = 0, where X ′ = µX ,
Y ′ = λY . By Lemma 15 we can use them to build our chart.
Corollary 17. If p is not an umbilical point of S ⊂ R3, there
is a coordinate system by lines of curvature around p.
Def.: Isometric and conformal maps, local isometries.
Proposition 18. Any surface (S, I) has isothermal charts.
Proof: Several proofs exist, few are elementary, and none easy...
§12. Ruled surfaces
ϕ(t, s) = α(s) + tv(s), v ∈ Xα, ‖v‖ = 1.
< v(s) > = geratrix line, α = directrix curve.
Remark 19. S ruled ⇒ K ≤ 0.
12
Examples:
• v = α′: Tangent surface to α
• α(s) = (0, 0, as), v(s) = (cos(s), sin(s), 0) ⇒ helicoid.
• If v ≡ v0 constant ⇒ cylinder over a plane curve. Hence,
we say that S is noncylindrical if v′ never vanishes.• α(s) = (cos(s), sin(s), 0), v = ±α′ + e3 ⇒ x2 + y2 − z2 = 1:
the hyperboloid of revolution is doubly ruled.
• α(s) = (s, 0, 0), v(s) = (0, 1, s) ⇒ xy = z, the hyper-
bolic paraboloid, that is also doubly ruled, since (s, t, st) =
te2 + s(1, 0, t).
Remark 20. Besides the plane, these are the only 2 doubly
ruled surfaces! How would you prove this??
Singularities of a ruled surface are contained in the striction curve:
Def.: For a noncylindrical ruled surface S, the striction curve
is given by σ(s) = α(s) + t(s)v(s) for which 〈σ′, v′〉 = 0 (i.e.,
t(s) = −〈α′, v′〉/‖v′‖2).Notice that the striction curve does not depend on the directrix
α. In particular, we can assume that σ = α, that is, 〈α′, v′〉 = 0.
§13. Minimal surfaces
The brachistochrone problem was formally posed by Johann
Bernoulli as a challenge (he knew the answer using the Fermat
Principle), but it appeared first in the Discorsi, of Galileo for
lines, and observed that there was a quicker non straight solution
(arguing then wrongly that the circle would be the fastest). Leib-
13
niz persuaded Bernoulli to extend the six month limit to solve the
challenge for foreign mathematicians to be able to participate.
Five more mathematicians solved the problem: Tschirnhaus, Ja-
cob Bernoulli, Leibniz, de L’Hopital, and... Isaac Newton, who
was teased by Bernoulli and Leibniz, and solved the problem in
one night. These solutions eventually lead to a general method by
Euler to solve these kind of problems: the calculus of variations.
When does a surface minimize the area for “close enough” sur-
faces?
Proposition 21. Let S ∈ R3 be a compact surface (with or
without boundary). Then, S is a critical point of the area
functional A(S) if and only if H = 0.
Proof: It is enough to consider normal variations it(p) = p +
tfN(p). Now, compute a′(0), where a(t) = Area(it(S)).
Exercise. Conclude the same for hypersurfaces adapting the proof using
that the volume element is given by√
det〈ϕui, ϕuj
〉 du1 ∧ · · · ∧ dun.
The Plateau problem: Find a minimal surface whose boundary is
a given closed curve. Douglas (1931) and Rado (1933) prove gen-
eral existence for arbitrary simple closed curves, but the surface
could have singularities. Osserman (1970) and Gulliver (1973): a
minimizing solution cannot have singularities. Regular solutions
may not exist.
CMC (hyper)surfaces.
14
§14. Intrinsic Geometry
Intrinsic objects of a Riemannian manifold are the ones that
only depend on the first fundamental form, i.e., invariant by
isometries: distance, angle, area, volume...
Cylinder ∼= plane ∼= cone (locally): they are intrinsically the same
thing, and hence the mean curvatureH is not an intrinsic concept.
If ϕ : U → M , ϕ′ : U → M ′ are charts such that gij = g′ij ⇒ϕ(U) ⊂ M and ϕ′(U) ⊂ M ′ are isometric.
Example: For a surface of revolution
ϕ(u, v) = (f (v) cos(u), f (v) sin(u), g(v))
we have E = f 2, F = 0, G = f ′2 + g′2. In particular, the
catenoid, where f (v) = a cosh(v) and g(v) = av for a > 0, has
E = G = a2 cosh2(v), F = 0. Now, change variables on the
helicoid
ϕ(u, v) = v(cos(u), sin(u), 0) + aue3,
v = a sinh(v), u = u to get
ϕ(u, v) = a(sinh(v) cos(u), sinh(v) sin(u), u),
that also has E = G = a2 cosh2(v), F = 0. Therefore, the
catenoid and the helicoid are locally isometric (but not globally).
This is a general phenomenon for minimal surfaces in R3: they
have a one parameter family of isometric deformations.
Def.: Isometries, conformal diffeomorphisms.
Existence of isothermal coordinates ⇒ any two surfaces are lo-
cally conformally equivalent.
15
§15. The Gaussian curvature in coordinates, part 2
Given a chart ϕ = ϕ(u1, u2) : U → S on a surface S ⊂ R3, we
have seen that K = (eg− f 2)/(EG−F 2). Let’s do this compu-
tation again in other way, by decomposing the second derivatives
on their tangent and normal components:
ϕij =∑
k
Γkijϕk + rijN.
The functions Γkij (that of course depend on ϕ) are called the
Christoffel symbols. (In our previous notation, r11 = e, r22 =
g, r12 = f ). Taking inner product with ϕi, we have:
Γ111E + Γ2
11F = 〈ϕ11, ϕ1〉 =1
2Eu1,
Γ111F + Γ2
11G = 〈ϕ11, ϕ2〉 = Fu1 −1
2Eu2,
· · ·that can be written as
(
E F
F G
)(
Γ111
Γ211
)
=
(
12Eu1
Fu1 − 12Eu2
)
and similarly for the other indexes. In other words, we have:
Proposition 22. The Christoffel symbols Γkij depend only on
the first fundamental form and its first derivatives.
Proof: Follows from the Koszul formula:
2〈ϕij, ϕk〉 = 〈ϕi, ϕk〉j + 〈ϕj, ϕk〉i − 〈ϕi, ϕj〉k.
16
Exercise. Show that for a surface of revolution around a curve α(v) =
(f(v), g(v)), it holds that Γ111 = Γ2
12 = Γ122 = 0,Γ2
11 = ff ′/(f ′2+g′2),Γ112 =
f ′/f,Γ222 = (ff ′′ + gg′′)/(f ′2 + g′2).
Now, we get relations that come from taking the tangent and
normal components of ϕrij = ϕrji and Nij = Nji (3×3 equations
if n = 2) that have the form
n∑
k=1
ckijrϕk + dkijrN = 0, ∀1 ≤ i, j, r ≤ n = 2. (2)
In particular, taking the ϕ2 component of ϕ112 = ϕ121 we obtain
the Gauss equation
K =1
E
(
(Γ212)1 − (Γ2
11)2 + Γ112Γ
211 + Γ2
12Γ212 − Γ2
11Γ222 − Γ1
11Γ212
)
.
We have proved the famous Gauss’ Egregium Theorem:
Theorem 23 (Theorema Egregium = “outstanding”). The
Gaussian curvature is an intrinsic concept (in fact, it depends
only on I, ∂I, and ∂2I). In particular:
K is invariant by (local) isometries.
Corollary 24. Kcatenoid(p) = Khelicoid(ξ(p)).
§16. The Codazzi-Mainardi equations
The other 5 tangential equations are other ways of writting the
Gauss equation, or give 0 = 0. But the normal components give
17
two more equations, called the Codazzi-Mainardi equations:
ev − fu = eΓ112 + f (Γ2
12 − Γ111)− gΓ2
11.
fv − gu = eΓ122 + f (Γ2
22 − Γ112)− gΓ2
12.
Application: The relative nullity integrate as straight lines.
These are the straight lines we ‘see’ in some ruled surfaces: K≡0.
Remark 25. In a coordinate system by lines of curvature of a
surface without umbilic points, Codazzi-Mainardi equations have
the form
ev =Ev
2
( e
E+
g
G
)
, gu =Gu
2
( e
E+
g
G
)
.
§17. Global application: The rigidity of S2 ⊂ R3
Lemma 26. Let p ∈ S ⊂ R3 regular such that K(p) > 0,
and p is a local maximum of k2 and a local minimum of k1(k1 ≤ k2). Then, p is umbilic.
Proof: Assume not, k1(p) < k2(p), and take a chart at p by lines
of curvature, 2H = k1 + k2 ⇒ k1 = e/E, k2 = g/G. Now,
by Remark 25 (Codazzi), ev = EvH , gu = GuH ⇒ E(k1)v =
ev− eEv/E = Ev(k2−k1)/2, and G(k2)u = −Gu(k2−k1)/2. In
particular, Ev(p) = Gu(p) = 0. But the Exercise in page 31 says
that −2KEG = Evv +Guu + (· · ·)Ev + (· · ·)Gu. Then, at p,
0>−2KEG=Evv+Guu=2(E(k1)vv−G(k2)uu)/(k2−k1) ≥ 0.
Theorem 27. (Liebman): If S ⊂ R3 is a regular connected
compact surface with K = constant ⇒ S is a round sphere
(i.e., S is umbilic).
18
Proof: Notice that K > 0 by compactness. Now, consider the
minimum of k1, and apply Lemma 26.
The same result follows with H = constant (Alexandrov). But a
weaker version of it (for K > 0) already follows exactly as above:
Theorem 28. If S ⊂ R3 is a regular connected compact
surface with K > 0 and H = constant ⇒ S is a round
sphere (i.e., S is umbilic).
§18. The Fundamental Theorem of surfaces in R3
We have seen that S ⊂ R3 ⇒ Gauss eq. (intrinsic) + Codazzi
equations (extrinsic). It turns out that there is no more informa-
tion, or, equivalently, the converse holds locally:
Theorem 29 (FTS: Bonnet). Let E,F,G, e, f, g be differ-
entiable functions on V ⊂ R2 with E,G > F 2 that sat-
isfy Gauss and Codazzi-Mainardi equations. Then, each q ∈V has a neighborhood q ∈ U ⊂ V and a diffeomorphism
ϕ : U → ϕ(U) ⊂ R3 such that E,F,G and e, f, g are the co-
efficients of the first and second fundamental forms of ϕ(U),
respectively, in the chart ϕ. In addition, if U is connected and
ϕ is another chart with the same E,F,G, e, f, g, then there is
a rigid motion T ∈ Iso(R3) such that ϕ = T ϕ.
Proof: (Sketch). For a chart ϕ, we define f := (ϕu, ϕv, N) :
V → GL(3,R) where N is given by (1). Hence, there are two
functions P,Q : V → R3×3 such that fu = fP , fv = fQ. Gauss
and Codazzi equations are precisely the integrability conditions
19
of this first order system of PDE: Pv − Qu = [P,Q] (Frobenius
Theorem). Integrating once more, we get ϕ, and it is easy to check
that it is a surface with the desired first and second fundamental
forms (for details, see R. Palais notes here).
Remark 30. Take a long time comparing this with the FTC.
§19. The covariant derivative: affine connections
We want to differentiate vector fields on our surface (submanifold)
S ⊂ R3. For this, we agree that, if L ⊂ R
m is a subspace and
v ∈ L, (v)L denotes the orthogonal projection of v onto L.
Definition 31. Given X ∈ X(S) and v ∈ TpS, we define the
covariant derivative of X in the direction v by
∇vX = (X∗p(v))TpS .
• ∇ is an intrinsic operator: depends only on Γkij;
• ∇ coincides with the usual derivative for S = Rn, since Γk
ij = 0;
• ∇vX is linear in v (tensorial!). So, we define for each Y ∈ X(S)
the vector field ∇YX ∈ X(S) by
(∇YX)(p) := ∇Y (p)X,
and this is tensorial in Y : ∇fYX = f∇YX ;
• ∇vX is a derivation in X : ∀f ∈ F(S), X ∈ X(S), v ∈ TpS,
∇vfX = v(f )X(p) + f (p)∇vX,
or, for Y ∈ X(S),
∇Y fX = Y (f )X + f∇YX.
20
a) In other words, we have:
∇ : X(S)× X(S) → X(S)
(Y,X) 7→ ∇YX , that is tensorial in Y and a derivation in X ;
b) ∇ is symmetric:
∇XY −∇YX = [X, Y ], ∀ X, Y ∈ X(M)
(it is enough to check for X = ∂i, Y = h∂j for any function h)
c) ∇ is compatible with the metric:
X〈Y, Z〉 = 〈∇XY, Z〉 + 〈Y,∇XZ〉, ∀ X, Y, Z ∈ X(M)
• Such an operator satisfying (a) + (b) + (c) always exists and is
unique by the Koszul formula:
2〈∇XY, Z〉 = X〈Y, Z〉 + Y 〈X,Z〉 − Z〈X, Y 〉
− 〈X, [Y, Z]〉 − 〈Y, [X,Z]〉 + 〈Z, [X, Y ]〉.In the realm of Riemannian Geometry,∇ is called the Levi-Civita
connection of (S, 〈 , 〉).
§20. Affine connections in vector bundles
Now, observe that, to have an affine connection (i.e., property (a)
only), all we need is the vector bundle structure on the second
variable, and not necessarily TS. Hence, we have:
Definition 32. Given a vector bundle π : E → M , an affine
connection in E is an R−bilinear operator
∇ : X(M)× Γ(E) → Γ(E),
21
(Y, ξ) 7→ ∇Y ξ, that is tensorial in Y and a derivation in ξ:
∇Y fξ = Y (f )ξ + f∇Y ξ,
∇fY ξ = f∇Y ξ,
∀ Y X(M), f ∈ F(M), ξ ∈ Γ(E).
Remark 33. Bump functions + local sections ⇒ affine connec-
tions are first order differential operators. In particular, they are
local: computations can be done in coordinates or local sections.
Therefore, if X, Y ∈ X(M) and ϕ : U → M is a chart, we write
on V = ϕ(U), X|V =∑
i xi∂∂ui
, Y |V =∑
i yi∂∂ui
, and since
∇ ∂∂ui
∂∂uj
=∑
k Γkij
∂∂uk
, in V we get for the Levi-Civita connection
(∇XY )|V =∑
k
∑
i
xi∂yk∂ui
+∑
ij
xiyjΓkij
∂
∂uk. (3)
Remark 33 also implies:
Proposition 34. Suppose E is a vector bundle with a con-
nection ∇. Then, for each smooth map f : N → M , there is
a unique pull-back connection ∇f on f ∗E satisfying that
∇fX(ξ f ) = ∇f∗Xξ, ∀ X ∈ X(N), ξ ∈ Γ(E).
Proof: Since connections are local objects, it is enough to do the
computation locally. But if ξi is a local frame of E in U ⊂ M ,
ξif is a local frame of f ∗E in V = f−1(U). So, if η ∈ Γ(f ∗E),
22
we write η on V as η|V =∑
i zi ξi f . Just by the definition of
a connection and its local nature, on V we get
∇fXη =
∑
i
(
X(zi)ξi f + zi∇f∗Xξi
)
. (4)
This implies the uniqueness of ∇f . But we can define ∇f locally
with (4): it is easy to check that ∇f defined this way is well
defined, and a connection.
§21. Affine connections along maps
If (M, 〈 , 〉) is a Riemannian manifold, the ONLY affine connec-
tion∇ on TM that we will consider is the Levi-Civita connection
of 〈 , 〉. If M is an Euclidean submanifold, we know how to con-
struct ∇ from the standard vector field derivative on Rm (that is
itself the Levi-Civita connection of Rm with the standard inner
product seen as a Riemannian metric).
As a particular case of Proposition 34, we have:
Proposition 35. Given f : N → (M, 〈 , 〉), there is a unique
affine connection ∇f(= f ∗∇) in f ∗(TM),
∇f : X(N)× Xf → Xf ,
called the affine connection along f , that satisfies
∇fX(Y f ) = ∇f∗XY, ∀X ∈ X(N), Y ∈ X(M).
In particular, for a curve α : I → M , we obtain:
• A notation: if X ∈ Xα,
X ′ := ∇d/dtX.
23
• We have the acceleration of α (intrinsic!):
α′′ := ∇d/dt α′,
• and the geodesic curvature of α (intrinsic!):
κg = καg := ‖α′′‖.
• “Compute derivatives using curves”: For any Z ∈ X(M) ⇒Z α ∈ Xα and
(Z α)′ = ∇d/dt(Z α) = ∇α′Z.
• For curves in a submanifold, α : I → M ⊂ Rm, if X ∈ Xα,
Z ∈ X(M), and v = α′(0) ∈ TpM , we have:
X ′ := ∇d/dtX =
(
dX
dt
)
TαM
∈ Xα
α′′ =
(
d2α
dt2
)
TαM
∈ Xα
∇vZ =
(
d
dt|t=0(Z α)
)
TpM
(5)
and the three curvatures of α are related by
κ2 = κ2g + κ2
n.
• Eq. (5) also implies that: if two submanifolds are tangent
along a curve α, their connections along α coincide.
• If M is a hypersurface, (v)TM = v−〈v,N〉N , and (5) becomes
∇vZ =d
dt|t=0(Z α)− 〈Apv, Z(p)〉N(p).
24
Exercise. A connection is compatible with a metric 〈 , 〉 ⇔ 〈V,W 〉′ =〈V ′,W 〉 + 〈V,W ′〉, for ever curve α and every V,W ∈ Xα (notice that
these are different “ ′”).
§22. Parallel transport
By (3), if we write a curve α locally as α = ϕ(α1, . . . , αn), and
Y =∑
i yi∂∂ui
α ∈ Xα, then
Y ′ =∑
k
y′k +∑
ij
α′iyjΓ
kij α
∂
∂uk α (6)
We say that Y ∈ Xα is parallel if Y ′ = 0. Since the last equation
is linear, the set of parallel vector fields along α, denoted by X‖α,
is a vector space. Also by this equation we easily see:
Proposition 36. Given a curve α : I → M , p = α(t0),
for every v ∈ TpM there exists a unique parallel vector field
µv ∈ Xα such that µv(t0) = v.
So, this map v 7→ µv is an isomorphism between TpM and X‖α. In
particular, if t is another point in I , we get a linear isomorphism
P αt0,t
: Tα(t0)M → Tα(t)M,
given by P αt0,t
(v) = µv(t). Notice that it depends smoothly on
everything.
Def.: This isomorphism P αt0,t
is called the parallel transport
along α between α(t0) and α(t).
25
Beware: along α!! It does depend on α, not just on α(t0) and
α(t) (in contrast to Rm).
Examples: M = Rm: usual. Meridian in S
2: Cylinder. Parallel
in S2: Cone ⇒ after a complete turn, the parallel transport does
not close ⇒ dependency on α.
Remark 37. By the previous exercise P αt0,t
are linear isometries.
Exercise. Prove that a connection is compatible with the metric 〈 , 〉 ⇔〈V,W 〉 is constant, for every curve α and every V,W ∈ X
‖α.
§23. Geodesics
Lemma 38. Given ϕ : U ⊂ R2 → M ⇒ ∇∂uϕv = ∇∂vϕu.
Proof: Use coordinates and the symmetry of∇ (it’s equivalent).
Proposition 39. A curve α parametrized by arc-length is a
critical point of the arc-length functional if and only if α′′ = 0.
Proof: Calculus of variations! :-)
Def.: A curve α with α′′ = 0 is called a geodesic.
Local and intrinsic concept ⇒ invariant by local isometries
Remark 40. Let α be a non-constant curve in (M, 〈 , 〉).• α is a geodesic ⇒ ‖α′‖ = constant ⇒ α is regular.
• α and α h are geodesics ⇔ h(t) = at + b.
• α is a geodesic ⇔ καg = 0.
• If α is a straight line segment in M ⊂ Rm ⇒ α is a geodesic
(once we parametrize it by arc length).
26
Examples:
• Great circles in round spheres are geodesics;
•More generally, meridians in surfaces of revolution are geodesics;
• Any helix inside cylinders;
• Given two points in a cylinder (not in the same parallel), there
are infinitely many geodesics joining them. But if we take a line
from the cylinder, we recover uniqueness (and existence).
• In R2 \ 0, (1, 0) and (−1, 0) have no geodesic joining them.
α is a geodesic ⇔ α′ ∈ X‖α. Then, by (6), we have that α =
ϕ(α1, . . . , αn) is a geodesic ⇔
α′′k = −
n∑
i,j=1
α′iα
′jΓ
kij α. (7)
This is the differential equation of geodesics, and implies:
Proposition 41. For every p ∈ M and every v ∈ TpM ,
there is ǫ > 0 and a unique geodesic γv : (−ǫ, ǫ) → M such
that γv(0) = p, γ′v(0) = v.
In fact, γv also depends smoothly on v.
§24. Geodesics in a surface of revolution
Let ϕ(u, v) = (f (v) cos(u), f (v) sin(u), g(v)) be a surface of revo-
lution with axis z and geratrix α(v) = (f (v), g(v)) parametrized
by arc length: ‖α′‖2 = f ′2 + g′2 = 1, f > 0. Notice that f is
the distance from the surface to the axis of revolution. Then, (7)
becomes
u′′ = −2f ′
fu′v′, v′′ = ff ′u′2. (8)
27
We get from this:
• Meridians are geodesics (very easy to see without this!)
• Parallels are geodesics ⇔ the distance function r = f to the
axis is critical: we can see this also geometrically with a picture.
Remark 42. The first equation in (8) can also be written
as f 2u′ = c = constant. Now, the angle θ ∈ [0, π/2] be-
tween the geodesic and the parallel that it intersects is given by
cos(θ) = |〈ϕu/‖ϕu‖, u′ϕu + v′ϕv〉| = |u′f |. Therefore, we have
the Clairaut relation:
r cos(θ) = constant,
where r = distance to the axis, θ = angle with parallel.
Let γ be a geodesic (p.b.a.l.) that is neither a parallel nor a merid-
ian. Then, f 2u′ = c 6= 0 is constant. But 1 = ‖γ′‖2 = f 2u′2+v′2
(by differentiating again, this implies the second equation in (8)).
So, v′ =√
f 2 − c2/f, u′ = c/f 2, and
u = c
∫
1
f√
f 2 − c2dv + u0.
In other words, we have integrated all the geodesic equations.
This is extremely rare, and we should thank the Clairaut relation,
that resumes the information about the geodesics.
Application: Let γ be a geodesic on the paraboloid of revolution
z = x2 + y2 that is not a meridian. Then, r cos(θ) = |c| 6= 0
⇒ θ grows with r, and θ = 0 (i.e. γ tangent to a parallel) only
at one point, the unique parallel r = |c| (limit of geodesics is a
28
geodesic, so it cannot accumulate over the parallel). Therefore, γ
intersects itself infinitely many times since it cannot be asymptotic
to a meridian (u = constant), since otherwise
u− u0 = c
∫
1
v
√
1 + 4v2
v2 − c2dv > c
∫
dv
v→ +∞.
§25. The covariant derivative on oriented surfaces
Let (S2, 〈 , 〉) oriented. Then, the oriented rotation of angle π/2
on TS is a skew symmetric orthogonal tensor J with J2 = −Id.
For w ∈ TS we use the notation w = Jw.
Let c be a regular curve in S2, and w ∈ Xc with ‖w‖ = 1. Then,
w′ = λw along c for some function λ =: [w′], called the algebraic
value of w′. In other words, [w′] = 〈w′, w〉. Accordingly, if c is
a curve in S2 parametrized by arc-length, we have the (oriented)
geodesic curvature of c,
κcg = [c′′].
Actually, for any manifold N and any map f : N → S2, if
w ∈ Xf is unitary, we have a 1-form [∇w] over N given by
[∇w](X) = 〈∇Xw,w〉.
Now, if w, e ∈ Xf are unitary ⇒ w = ae + be, a2 + b2 = 1 ⇒
Lemma 43. With the notations above, assume N is simply
connected, and fix p ∈ N . If cos ξ0 = a(p) and sin ξ0 = b(p),
then there is a unique differentiable function ξ = ∢(w, e) :
N → R such that cos ξ = a, sin ξ = b, and ξ(p) = ξ0.
29
Proof: Define σ ∈ Ω1(N) by σ(X) = aX(b) − bX(a). Since
a2 + b2 = 1 we easily check that σ is closed, hence exact. Define
now ξ by dξ = σ, ξ(p0) = ξ0. The lemma follows simply by
differentiating (a−cos ξ)2+(b−sin ξ)2 = 2−2(a cos ξ+b sin ξ).
Def.: Given w and e ∈ Xf unitary, the differentiable function
∢(w, e) is called a determination of the angle between w and e.
For a non vanishingX ∈ X(S), we set ∢(w,X) := ∢(w, X‖X‖f ).
Lemma 44. If f : N → S with N simply connected, and
w, e ∈ Xf are unitary ⇒ [∇w]−[∇e] = dξ, where ξ = ∢(w, e).
Proof: Just compute [∇w] using that w = cos(ξ) e + sin(ξ) e.
Remark 45. In particular, if α is parametrized by arc-length
and w ∈ X‖α ⇒ κα
g = [α′′] = ξ′, where ξ = ∢(w, α′). Therefore:the geodesic curvature of a curve is the rate of change of the
angle of its tangent and a parallel vector field along it.
From now on, ϕ : U → S will be an orthogonal oriented chart of
S, and N any simply connected manifold.
Lemma 46. Let f : N → ϕ(U) ⊂ S, and write f (x) =
ϕ(u(x), v(x)) for some u, v : N → R. If w ∈ Xf is unitary,
[∇w] =1
2√EG
(Gudv − Evdu) + dξ, where ξ = ∢(w,ϕu).
Proof: Define the vector fields V = ϕu/√E ∈ Xϕ (unitary), and
e(x) = V (u(x), v(x)) ∈ Xf . By Lemma 44, [∇w]−dξ = [∇e].
But
[∇V ](∂u) = 〈∇∂u(ϕu/√E), ϕv/
√G〉 = 〈∇∂uϕu, ϕv〉/
√EG
= −〈∇∂uϕv, ϕu〉/√EG = −Ev/2
√EG,
30
by Lemma 38. Analogously, [∇V ](∂v) = Gu/2√EG. The lemma
follows from ∇Xe = ∇X(V (u, v)) = ∇X(u)∂u+X(v)∂vV .
Corollary 47. If α(s) = ϕ(u(s), v(s)) is a curve p.b.a.l.,
καg =
1
2√EG
(Guv′ − Evu
′) + ξ′, where ξ = ∢(α′, ϕu).
Exercise. In an orthogonal chart, −2K√EG =
(
Gu√EG
)
u+
(
Ev√EG
)
v.
§26. The local Gauss-Bonnet Theorem
Gauss: Geodesic triangles and excess.
In this section, S will be an oriented surface, charts will be com-
patible with its orientation, and α : I → S will be a simple,
closed, piecewise regular curve with vertices α(t1), . . . , α(tn)⇒ oriented external angle θi ∈ [−π, π] at the vertex α(ti).
Let ϕ : U ∼= D2 := x ∈ R2 : ‖x‖ < 1 → S be a chart, α(I) ⊂
ϕ(U), and ξi = ∢(α′|[ti,ti+1], ϕu) : [ti, ti+1] → R (tn+1 := t1).
Proposition 48. (Turning tangents) With these notations,
n∑
i=1
(ξi(ti+1)− ξi(ti)) +
n∑
i=1
θi = ±2π,
where the RHS sign depends on the orientation of α.
Proof: The LHS is the total change of the angle between α′
and ϕu. Since α is closed, this is 2kπ, for some integer k, hence
invariant under homotopies of α. Since α is simple, k = ±1.
31
Def.: We say that a compact region R ⊂ S is simple if R ∼= D2
and ∂R is the trace of a closed simple piecewise regular curve α,
that we orient in such a way that α′ points to the interior of R.
Theorem 49. (Gauss-Bonnet; local) Let R ⊂ S be a simple
region contained in the image of an orthogonal oriented chart
ϕ : U ∼= D2 → S, and let α : I → ∂R oriented p.b.a.l. with
vertices α(si) and external oriented angles θi. Then,
∑
i
∫ si+1
si
καg +
∑
i
θi +
∫
R
K = 2π.
Proof: Write α(s) = ϕ(u(s), v(s)). Integrating Corollary 47 and
using Green’s Theorem and the previous exercise we obtain∑
i
∫ si+1
si
καg ds −
∑
i
(ξi(si+1)− ξi(si))
=∑
i
∫ si+1
si
(
Gu
2√EG
v′ − Ev
2√EG
u′)
ds
=
∫
ϕ−1(R)
((
Gu
2√EG
)
u
+
(
Ev
2√EG
)
v
)
dudv
= −∫
ϕ−1(R)
(K ϕ)√EG dudv = −
∫
R
K.
Now the result follows from Proposition 48.
Remark 50. Let α : I → S a regular simple closed curve
parametrized by arc-length, and w ∈ X‖α unitary. Then,
0 =
∫ ℓ
0
[w′] =
∫ ℓ
0
(
Gu
2√EG
v′ − Ev
2√EG
u′)
ds +
∫ ℓ
0
ξ′
32
= −∫
R
K + ξ(ℓ)− ξ(0) = ∆ξ −∫
R
K,
where ξ = ∢(w,ϕu). So, limR→p∆ξA(R) = K(p), and we conclude:
The parallel transport is (locally)
independent of the path α ⇔ K ≡ 0.
§27. Gauss-Bonnet: What’s happening?
By Lemma 46 and the previous exercise, if w is any unit vector
field on an open set V ⊂ S, then on V it holds that
−KdA = d [∇w]. (9)
(Notice that Lemma 44 explains why the LHS does not depend
on w). Thus, if R is a small simple region with smooth boundary
parametrized by an oriented α, and ξ = ∢(α′, w), we get from
Lemma 44 and Stokes’ theorem applied to (9) that∫
R
K = −∫
∂R
[∇w] = −∫ ℓ
0
[(wα)′] =∫ ℓ
0
(ξ′−καg ) = 2π−
∫ ℓ
0
καg .
Now, we get Theorem 49 from this by approximating R with
domains with regular boundaries.
§28. The global Gauss-Bonnet Theorem
Recall: Triangulations and the Euler characteristic of manifolds
χ(M) (see Theorem 33 in our Analysis on Manifolds notes here).
Example: If n-torus := S2+n handles ⇒ χ(n−torus) = 2−2n.
33
For compact connected surfaces it holds that:
χ(S) = χ(S ′) ⇔ S is (diffeo)homeomorphic to S ′ (!!)
If S is orientable ⇒ 4 − χ(S) ∈ 2N. Therefore, by Jordan’s
Theorem, the only compact regular surfaces in R3 are the n-tori
(up to diffeomorphism). Thus, the number g of “handles” of S,
g := 2−χ(S)2 ∈ N0, is called the genus of S.
Def.: A region R ⊂ S is regular if it is compact and ∂R is a
disjoint union of simple closed piecewise differential curves.
Theorem 51. (Gauss-Bonnet; global) Let R be a regular re-
gion of an oriented surface S, and let ∂R = ∪ki=1Ci positively
oriented with external angles θ1, . . . , θm. Then,k
∑
i=1
∫
Ci
καg +
m∑
j=1
θj +
∫
R
K = 2πχ(R).
Proof: Take a fine triangulation of R, T = Ti, . . . , TF, suchthat each triangle lies in a simple orthogonal coordinate system,
and orient each triangle Ti according to the orientation of S. Let
E be the number of edges and V the number of vertices of the
triangulation, Ei and Ee the number of internal and external
edges, respectively, and Vi and Ve the number of internal and
external vertices, respectively. In addition, we have Ve = Vec+Vet,
where Vec is the number of external vertices that are vertices of
Ci, while Vet is the number of vertices on regular points of ∂R.
Adding the local Gauss-Bonnet Theorem 49 for each Ti givesk
∑
r=1
∫
Cr
καg +
F∑
r=1
3∑
j=1
θrj +
∫
R
K = 2πF,
34
where θrj are the 3 oriented external angles of Tr, since the
internal edges of each triangle get opposite orientations. Call
βrj := π − θrj the internal angles of the triangle Tr. Thus,
3πF −∑
rj
θrj =∑
rj
βrj = 2πVi + πVet +
m∑
l=1
(π − θl).
Now, since each Cr is closed, Ee = Ve. Moreover, by counting
each triangle edges we get 3F = 2Ei + Ee. And, since m = Vec,
∑
rj
θrj−m∑
l=1
θl=π(2Ei+Ee−2Vi−Vet−m) = π(2E−2V ).
Corollary 52. The total curvature of a compact oriented
surface is a purely topological invariant:∫
S K = 2πχ(S)
Corollary 53. Local Gauss-Bonnet for simple regions.
Corollary 54. S compact orientable surface with K ≥ 0 ⇒S ∼= S
2 or S ≡ S1 × S1. If, in addition, S ⊂ R3 ⇒ S ∼= S
2.
Corollary 55. S orientable with K ≤ 0 ⇒ 2 geodesics do
not enclose a simple region. In particular, a closed geodesic
or a geodesic loop do not enclose a simple region.
Corollary 56. S ∼= cylinder with K < 0 ⇒ S has at most
one closed geodesic (compare with the flat cylinder).
Corollary 57. S compact, K > 0 ⇒ two closed geodesics
intersect (compare with K ≥ 0: a flat cylinder with two caps).
35
Corollary 58. (Gauss) The excess in the internal angles of
a geodesic triangle is equal to its total curvature. ⇒
Corollary 59. H2: the fifth Euclid axiom is independent.
§29. Application: Total Index of a vector field
Def.: The index I(p) of isolated singularity p of a vector field
X ∈ X(S) is the integer given by
2πI(p) =
∫ ℓ
0
τ ′ = τ (ℓ)− τ (0) = ∆τ,
τ = ∢(X,ϕu) α, for a small curve α around p. By Remark 50,∫
R
K − 2πI(p) = ∆(ξ − τ ) = ∆(∢(w,X)),
that does not depend on ϕ (it is also independent of α).
Therefore, if X is a vector field in a compact oriented surface
with isolated singularities p1, . . . , pn (a generic property), by
choosing a smart triangulation we get∫
S K − 2π∑
i I(pi) = 0,
since the boundaries of the triangles appear twice with opposite
orientations. By Corollary 52 we thus have for the total index∑
i I(pi) of X :
Theorem 60. (Poincare-Hopf) The total index of a vector
field in a compact oriented surface S with isolated singulari-
ties is equal to the Euler characteristic of S:∑
i I(pi) = χ(S).
36
PART IIPrerequisites (part II): Analysis on manifolds, Stokes and de Rham cohomology.
Bibliography: [KN] Vol.II, Ch.12; [S] Vol.V Ch.13; [MS]
§30. Fiber and principal bundles ([KN], Vol. I, Ch. 1.4, 1.5)
Lie group G; left-invariant vector fields ∼= Lie algebra g = TeG:
Vv(g) := (Lg)∗ev.
Proposition 61. v ∈ g ↔ one parameter subgroup βvt of G.
Proof: If ξt(g) the flux of Vv, βvs = ξs(e) ⇒ βv
0′ = v, and
Vv(βvsβ
vt ) = (Lβvs )∗βvt Vv(β
vt ) = (Lβvs )∗βvt β
vt′ ⇒ βv
sβvt is an integral
curve of Vv passing at βvs for t = 0 ⇒ βv
sβvt = βv
s+t.
Exercise. Show that the flux of Vv is (g, s) 7→ gβvs = Rβv
s(g).
Representations; Adjoint representation: adG : G → End(g)
Lemma 62. If v ∈ g, [v, · ] = (adG)∗e(v) =dds |s=0
adβvs .
Proof: By the exercise in §11, since Rβvs (g) is the flux of Vv,
[v, w] = [Vv, Vw](e) = lims→0
1
s((Rβv−s
)∗Vw(βvs )− Vw(e))
= lims→0
1
s((Rβv−s
)∗(Lβvs )∗ew − w) =d
ds |s=0
adβvs (w).
Group actions on manifolds: R : E ×G → E; free actions
Fiber bundles F → Eπ→ B with typical fiber Ep
∼= F , total
space E and base B
Transition functions: ξUV : U ∩ V → Diff (F )
Structure group of a fiber bundle: G-bundles:
37
Def.: A G-bundle is a fiber bundle F → Eπ→ B together with
a left action on F by G, ρ : G×F → F , such that the transition
functions are given through ρ, that is, there are ξUV : U∩V → G
and ξUV (x)(f ) = ρ(ξUV (x), f ).
(transition functions act on the left)
Exercise. Show that a rank k vector bundle is a Gl(k,R)-bundle.
Exercise. Show that the pull-back and Whitney sum of vector bundles is
a vector bundle.
Exercise. Show that the tangent bundle of S3 is trivial.
Example: Sphere bundles
Bundle maps, bundle isomorphism
Def.: A principal G-bundle is as a G-bundle π : E → B
with fiber G where the structure group acts on the fibers by left
multiplication.
Remark 63. By the associativity of the group, the right mul-
tiplication by G on the fiber commutes with the action of the
structure group (left multiplication). So we get an invariant right
action by G on E. This action preserves the fibers of E and acts
freely and transitively on them. Actually, principal bundles can be
defined with the use of a free right action transitive on the fibers.
Examples: Product B × G. If (S, 〈 , 〉) is an oriented surface,
the unit tangent bundle T1S is an S1-principal bundle. The (or-
thonormal) frame bundle of a rank k vector bundle is a principal
(O(k)-bundle) GL(k,R)-bundle. Projective spaces.
38
Exercise. Show that all these are principal bundles.
Exercise. Show that the S1 and S
3 Hopf bundles are principal bundles.
Exercise. Show that the lens spaces S3/Zk for Zk ⊂ S1 are circle bundles
over S2.
Fact: Principal G-bundles “generate” all G-bundles, via the as-
sociated bundles (we will see this in §34).(Cross-)sections
Local sections ⇔ (equivariant) local trivializations
Obs.: A principal bundle E has a global section ⇔ E is trivial.
Compare to vector bundles, that always have 0 as a global section.
§31. Connections ([KN], Vol. I, Ch. 2.1, 2.5)
Def.: The vertical subbundle V ⊂ TE is the vector bundle over
B given by V = Ker π∗.
Fundamental vec. fields: If ξp(g)=Rg(p)=pg, v∈g 7→v∗∈X(V),
v∗(p) :=d
dt |t=0
(pβvt ) = ξp∗e(v).
Notice that (t, p) 7→ pβvt = Rβvt
(p) is the flux of v∗.
Exercise. Show that v∗ ξp = ξp∗ Vv, i.e., v∗ ξp∼ Vv for all p ∈ E. In
particular, [v∗, w∗] = [v, w]∗, i.e., v 7→ v∗ is an algebra homomorphism.
Def.: A connection on a principal bundle E is a differentiable
map that assigns to each x ∈ E a subspace Hx ⊂ TxE such that:
• TE = V ⊕H;
39
• H is G-invariant: Hpg = (Rg)∗p(Hp), ∀ p ∈ E, g ∈ G.
Obs.: TE = V ⊕H ⇒ π∗p|Hp : Hp → Tπ(p)B is isomorphism
Def.: A type adG k-form on E is a g-valued k-form σ : TE ×· · · × TE → g that satisfies
R∗gσ = adg−1 σ, ∀ g ∈ G.
Def.: A principal connection on E is a type adG 1-form w :
TE → g such that H = Kerw, and w(v∗) = v for all v ∈ g.
Exercise. There is a 1-1 correspondence between the two type of connec-
tions (w is of type adG ⇔ H is G-invariant).
Exercise. Principal connections always exist (use partitions of unitiy).
H and V components: X = Xh +Xv. Define h(X) := Xh.
Obs.: GivenX ∈ X(B), there is a uniqueX∗ ∈ X(E), called the
lift of X , such that X ∈ H and π∗(X∗) = X π (i.e., X∗ π∼ X).
Moreover, the lift is G-invariant (i.e., X∗ Rg∼ X∗), and, conversely,every G-invariant horizontal vector field on E is a lift.
Exercise. [X∗, Y ∗]h = [X, Y ]∗.
Lemma 64. If v∗ is a fundamental vector field and Y ∈ H,
[v∗, Y ] ∈ H. If, in addition, Y is a lift, then [v∗, Y ] = 0.
Proof: By exercise in §11, [v∗, Y ] = limt→01t ((ξ−t)∗Y ξt−Y ),
where ξt = Rβvtis the flux of v∗. Thus, ((ξ−t)∗Y )(ξt(p)) ∈ Hp.
Def.: A k-form σ on E is horizontal if V ⊆ Ker σ, i.e., if it
vanishes if one of the entries is vertical.
40
Lemma 65. Let σ be a horizontal k-form in E that is G-
invariant, i.e., R∗gσ = σ. Then, σ projects to σ, i.e., there
exist a k-form σ in B such that σ = π∗σ.
Proof: Define σ(Z1, . . . , Zk)(x) = σ(Z∗1 , . . . , Z
∗k)(p), where p ∈
π−1(x). This is independent of p since the lift is G-invariant.
If σ is a k-form on E, we define the k-form σh by
σh(X1, . . . , Xk) := σ(Xh1 , . . . , X
hk ).
If σ is of type adG, then so dσ and σh are (because (Rg)∗ h =
h (Rg)∗). In addition, σh is always horizontal. So:
Def.: The horizontal (k + 1)-form Dσ = (dσ)h is called the
exterior covariant derivative of σ (it is of type adG if σ is).
Lemma 66. If σ projects to σ, then Dσ = dσ.
Proof: The obvious: dσ(Y0, . . . , Yk) = d(π∗σ)(Y0, . . . , Yk) =
dσ(π∗Y0, . . . , π∗Yk)=dσ(π∗Y h0 , . . . , π∗Y
hk )=dσ(Y h
0 , . . . , Yhk ).
§32. The curvature of a principal connection
Def.: If w is a connection form on E, the horizontal 2-form Dw
is of type adG and is called the curvature form of E.
From now on, Ω := Dw will be the curvature 2-form of (E,w).
Proposition 67. (Structure equation) Ω = dw + [w,w], i.e.,
Ω(X, Y ) = dw(X, Y ) + [w(X), w(Y )], ∀X, Y ∈ TE.
41
Proof: If X, Y ∈ H, follows from the definition of D. If
X = A∗ ∈ V is fundamental, Ω(A∗, ·) = 0 because it is verti-
cal. Now, if B∗ ∈ V is fundamental, dw(A∗, B∗) = A∗(w(B∗))−B∗(w(A∗))−w([A∗, B∗])=A∗(B)−B∗(A)− [A,B]=−[A,B] =
−[w(A∗), w(B∗)]. Now, if Y ∈ H, [w(A∗), w(Y )] = 0 and
dw(A∗, Y ) = −w([A∗, Y ]) = 0 by Lemma 64.
Proposition 68. (Bianchi’s identity) DΩ = 0.
Proof: We need to check that dΩ = d([w,w]) = 0 for 3 horizontal
vector fields, but this is immediate from H ⊂ Kerw.
Lemma 69. If σ is a horizontal 1-form of type adG,
Dσ = dσ + [σ, w] + [ω, σ].
Proof: The only nontrivial case is for v∗ ∈ V fundamental and
Y ∗ ∈ H a lift. ButDσ(v∗, Y ∗) = 0, and dσ(v∗, Y ∗) = v∗(σ(Y ∗))since [v∗, Y ∗] = 0 by Lemma 64. Since Y ∗ is G-invariant, by
Lemma 62,
v∗(σ(Y ∗))(p) = σ(Y ∗(pβvs ))
′ = σ((Rβvs )∗pY∗(p))′
= ((R∗βvsσ)(Y ∗(p)))′ = (adβ−v
s(σ(Y ∗(p))))′
= [−v, σ(Y ∗(p))] = −[w(v∗), σ(Y ∗)](p).
§33. Weil homomorphism ([KN], Vol. II, Ch. 12.1)
Let Ik(G) be the set of symmetric k-multilinear maps over g, f :
g× · · ·× g → R and adG-invariant, i.e. f (adgX1, . . . , adgXk) =
f (X1, . . . , Xk). This is a vector space, and I(G) = ⊕∞k=0I
k(G)
is a graded algebra with the natural product (fg)(t1, . . . , tk+s) =1
(k+s)!
∑
σ f (tσ1, . . . , tσk)g(tσk+1, . . . , tσk+s
).
42
Now, let E be a principal G-bundle with principal connection
1-form w and curvature 2-form Ω. For f ∈ Ik(G), we define the
2k-form f (Ω) = f (Ω, . . . ,Ω) on E by
f (Ω)(X1, . . . , X2k)
=1
(2k)!
∑
σ
sign(σ)f (Ω(Xσ1, Xσ2), . . . ,Ω(Xσ2k−1, Xσ2k)).
Theorem 70. (A.Weil) For each f ∈ Ik(G), the (2k)-form
f (Ω) ∈ Ω2k(E) projects to a unique closed (2k)-form f (Ω) ∈Ω2k(B). Moreover, its cohomology class
ωf = [f (Ω)] ∈ H2k(B)
is independent of the choice of the connection, and ω: I(G) →H∗(B) is an algebra homomorphism, called Weil homomorp..
Proof: Since Ω is horizontal by definition, so is f (Ω). Since
Ω is of type adG and f is adG-invariant, f (Ω) is G-invariant:
R∗g(f (Ω))=f (Ω). By Lemma 65 f (Ω) projects: f (Ω) = π∗f (Ω).
Proposition 68 says that DΩ = 0, and hence D(f (Ω)) = 0. By
Lemma 66, f (Ω) is closed, and so is f (Ω) since π∗ is onto.
For the second part, take w1 and w2 two principal connections
on E, and define wt := w0 + t(w1 − w0). Obviously, wt and
α = w1 − w0 are also of type adG, and α(V) = 0. Let Dt and
Ωt be the exterior covariant differentiation and curvature form of
wt, respectively. By Proposition 67, Ωt = Dtwt = dwt + [wt, wt],
so, by Lemma 69, ddtΩt = Dtα. Therefore, by Proposition 68,
d
dtf (Ωt) = kf (Dtα,Ωt, . . . ,Ωt) = kDt(f (α,Ωt, . . . ,Ωt))
= kd(f (α,Ωt, . . . ,Ωt)),
43
where the last equality follows from Lemma 65 and Lemma 66:
since both α and Ωt are horizontal and of type adG, and f is adG-
invariant, so βt := f (α,Ωt, . . . ,Ωt) is horizontal and G-invariant.
But then βt also projects to a (k−1)-form on B, and so does Φ =
k∫ 1
0 βtdt. We conclude from the above that dΦ = f (Ω1)−f (Ω0)
also projects, and thus f (Ω1)− f (Ω0) = dΦ is exact.
It’s easy to check that ω is an algebra homomorphism (exercise).
Remark 71. Notice that the homology class is in the base B,
not in the total space E!!
Def.: The class ωf is called the characteristic class of E associ-
ated to f , that, by Theorem 70, depends only on the isomorphism
class of the bundle, and not on the choice of the connection.
§34. Associated bundles
Take a G-bundle F → E → B with its G-action ρ : G×F → F
and transition functions ξUV : U ∩ V → G. If F ′ is another
manifold where G acts via ρ′ : G × F ′ → F ′, we can construct
another G-bundle F ′ → E ′ → B associated to the original one
by using the same transition functions ξUV but simply changing
F by F ′ and ρ by ρ′ (see here for details).In particular, we can take F ′ = G and ρ′ = left multiplication to
get the G-principal bundle associated to the original one.
This allows us to define the characteristic classes of any G-bundle
as the characteristic classes of its associated principal bundle.
In particular, for a real vector bundle of rank k its associated
Gl(k,R)-principal bundle is nothing but its frame bundle.
44
§35. The shortcut for vector bundles ([MS], Appendix C)
Let’s show a much more direct approach for vector bundles. No-
tice that, since G = Gl(n,R) ⊂ Rn×n is open, g = R
n×n.
Let Rn → Pπ→ M be a real rank n vector bundle over a manifold
M , and ∇ an affine connection on P . Given a local frame e =
ξ1, . . . ξn of π−1(U) ∼= U × Rn, U ⊂ M , write
∇Xξj =∑
i
Γij(e)(X)ξi.
So, ω(e) = (Γij(e)) are g-valued 1-forms on U that determine ∇.
Exercise. For g:U → G, ω(eg) = g−1dg + adg−1(ω(e)).
Define the g-valued curvature 2-form Ω(e) of ∇ by
∇X∇Y ξj −∇Y∇Xξj −∇[X,Y ]ξj =∑
i
Ωij(e)(X, Y )ξi.
It is easy to check that Ω satisfies the structure equation (com-
pare with Proposition 67):
Ω(e) = dω(e) + [ω(e), ω(e)],
the Bianchi identity (compare with Proposition 68):
dΩ(e) = [Ω(e), ω(e)]
(i.e., [Ω, ω]ij =∑
k(Ωik∧wk
j −wik∧Ωk
j ), or [Ω, ω](X1, X2, X3) =1
2
∑
σ∈S3[Ω(Xσ1
, Xσ2), ω(Xσ3
)])
and changes as Ω(eg) = g−1Ω(e)g = adg−1(Ω(e)) (exercises).
Thus, if f is an adG-invariant homogeneous polynomial of degree
k as before, f (Ω) is a well defined (i.e. independent of the local
frames e) and thus global (2k)-form on Mn. In addition, f (Ω) is
closed (easy exercise using Bianchi), so [f (Ω)] ∈ H2k(M).
45
Now, [f (Ω)] does not depend on the affine connection. If ∇1,∇2
are two affine connections on P , then ∇t = (1 − t)∇0 + t∇1
is also an affine connection on P . Consider the projection π1 :
M × R → M and it : M → M × R, it(x) = (x, t). The
connection ∇ = π∗1∇t is an affine connection on the vector bundle
π∗1(P ) → M ×R, so the corresponding f (Ω) is closed on M ×R.
But i∗ǫ(f (Ω)) = f (Ωǫ), for ǫ = 0, 1 and, since i0 and i1 are
homotopic, [f (Ω0)] = [f (Ω1)].
35.1 Affine connections ⇔ principal connections
Let’s see that the two constructions agree. Given P the vector
bundle above, its frame bundle of G → F(P )π→ M is a principal
G-bundle, a trivializing neighborhood of which is F(π−1(U)) ∼=U × G. If w is a principal connection on F(P ), and e : U ⊂M → π−1(U) ⊂ F(P ) is a local section, the equation
ω(e) = e∗w (10)
relates w with the affine connection form ω. Then, check that
Ωω(e) = e∗Ωw, and so the forms f (Ωw) project precisely to f (Ωω):
π∗(f (Ωω)) = π∗(f (e∗Ωw)) = π∗e∗(f (Ωw)) = f (Ωw), (11)
where for the last equality we used that f (Ωw) projects.
Exercise. If ω and w are forms related by (10), then ω is well defined,
and it is a principal connection ⇔ w is an affine connection (form).
Now, put a Riemannian metric on P and work with the orthonor-
mal frame bundle. If e is an orthonormal frame and ∇ is com-
patible we get Γij(e) = −Γj
i (e), so ω(e) is still a g-valued 1-form
on U but now for g = o(n), and we play as before.
46
In particular, all this holds for P = TM when M is Riemannian.
35.2 Gauss-Bonnet: What’s REALLY happening??
We can now understand more deeply the Gauss-Bonnet theorem:
1. TS as an oriented vector bundle. If S is a oriented Rieman-
nian surface, its Levi-Civita connection (form) of an orthonormal
oriented frame e = e1, e2 is a standard 1-form since so(2) = R,
and is given by ω(e) = −[∇e1]. Its curvature form is Ω(e) =
dω(e) = KdA by (9). Taking f (t) = t, Ω = f (Ω(e)) = KdA is
a well-defined and global closed 2-form, whose cohomology class
is independent of the compatible affine connection, in particular,
independent of the metric, and so is∫
K.
Now, if e1 is globally defined but in a finite set pi the curvature
form is then exact almost everywhere. We remove ǫ-small disks
Dǫi around each pi and we use Stokes and Theorem 60 to get∫
K=limǫ→0
∫
M\∪iDi
Ω =∑
i
limǫ→0
∫
∂Dǫi
−[∇e1] = 2π∑
i
I(pi) = 2πχ(S).
2. T1S as a SO(2) = S1-principal bundle. The S
1 action
on T1S is given by uθ = cos(θ)u + sin(θ)u. We can choose as a
principal connection 1-form w(u∗(X)) = −[∇u](X), where u is
a section of T1S and X ∈ TS. Since S1 is abelian, the curvature
2-form of w is Ω = dw and projects to a closed two form on S
whose cohomology class does not depend on the metric. Indeed,
u∗Ω = u∗dw = du∗w = dω(u, u) = Ω(u, u) = Ω
does not depend on u and therefore π∗Ω = Ω.
47
§36. Invariant polynomials ([KN], Vol. II, Ch. XII.2)
Let P k(V) be the homogeneous polynomial functions on the (fi-
nite dimensional) vector space V of degree k (polynomial by tak-
ing a basis), and P (V)=⊕∞k=0P
k(V) the natural algebra of poly-
nomial functions. Let Sk(V) be the set of symmetric k-multilinear
functions on V, with S(V)=⊕∞k=0S
k(V) its natural commutative
algebra.
Proposition 72. (Polarization). The map τ :S(V) 7→ P (V)
given by τ (h)(t)=h(t, . . . , t) is an algebra isomorphism.
Proof: If ξ1, . . . , ξn is a basis of V∗, f ∈ P k(V ) can be written
as∑
fi1...ikξi1 · · · ξik , for some fi1...ik ∈ R symmetric in the in-
dexes. The function Φ(f )(t1, . . . , tk) =∑
fi1...ikξi1(t1) · · · ξik(tk)
is the inverse of τ (exercise).
Exercise. If G ⊂ L(V) is a subgroup, the isomorphism above induces an
isomorphism between the G-invariant subalgebras SG(V) and PG(V).
Corollary 73. I(G) ∼= P (G), where P (G) are the adG-
invariant polynomial functions in g.
36.1 The unitary group: U(n) = X ∈ Cn×n : XX
t= I.
Its Lie algebra is u(n) = A ∈ Cn×n : A
t= −A. If A ∈ u(n),
det(λI+i
2πA) = λn−σ1(A)λ
n−1+σ2(A)λn−2−· · ·+(−1)nσn(A).
Then, the polynomial functions σi are adU(n)-invariant. In fact,
if it1, . . . , itn are the eigenvalues of A, then σi(A) is the i-th
symmetric function on t1, . . . , tn. And these are all:
48
Proposition 74. The polynomial functions σ1, . . . , σn are
adU(n)-invariant, algebraically independent, and generate (as
algebra) PU(n)(u(n)).
Proof: See Theorem 2.5 in Kobayashi-Nomizu, Vol 2, Cap. XII.
Corollary 75. The characteristic classes ck(E) := ωσk ∈H2k(B), 1 ≤ k ≤ n, generate all the characteristic classes
of an U(n)-principal bundle U(n) → E → B as an algebra.
They are called the Chern classes of the bundle.
36.2 The orthogonal group: O(n) = X ∈ Rn×n : XX t = I.
Its Lie algebra is o(n) = A ∈ Rn×n : At = −A. Define
det(λI − 1
2πA) = λn + p1(A)λ
n−2 + p2(A)λn−4 + · · · + · · ·
Then, the polynomial functions p1, . . . , p[n/2] are adO(n)-invariant.
In fact, if ±it1, . . . ,±it[n/2] are the eigenvalues of A (besides
the 0 if n is odd), then pi(A) is the i-th symmetric function on
t21, . . . , t2[n/2]. And these are all:
Proposition 76. The polynomial functions p1, . . . , p[n/2] are
adO(n)-invariant, algebraically independent, and generate (as
algebra) PO(n)(o(n)).
Proof: See Theorem 2.6 in Kobayashi-Nomizu, Vol 2, Cap. XII.
Corollary 77. The characteristic classes pk(E) := ωpk ∈H4k(B), 1 ≤ k ≤ [n/2], generate all the characteristic classes
of an O(n)-principal bundle as an algebra. They are called
the Pontrjagin classes of the bundle.
49
36.3 Special orthogonal group: SO(n)=X ∈ O(n): det(X) = 1.
Being SO(n) the connected component of O(n) containing the
identity, their Lie algebras coincide so(n) = o(n), and the situa-
tion is very similar to that of O(n). However, for n = 2m even,
there is a (unique up to sign) SO(n) invariant homogeneous poly-
nomial function pf such that pf 2 = pm, called the pfaffian. In
terms of matrixes, pf (A)2 = det(A), and is given by
pf (A) =1
2mm!
∑
σ∈S2m
sign(σ)m∏
i=1
aσ(2i−1)σ(2i). (12)
Hence, we have:
Proposition 78. For n = 2m − 1 (resp. n = 2m) the
polynomial functions p1, . . . , pm−1 (resp. p1, . . . , pm−1, pf) are
adSO(n)-invariant, algebraically independent, and generate (as
algebra) PSO(n)(so(n)).
Proof: See Theorem 2.7 in Kobayashi-Nomizu, Vol 2, Cap. XII.
Corollary 79. The characteristic classes pk(E) := ωpk ∈H4k(B), 1 ≤ k ≤ [n−1
2 ], together with e(E) := (2π)−n/2ωpf ∈Hn(B) if n is even, generate all the characteristic classes of
an SO(n)-principal bundle as an algebra. The classes pi(E)
are called the Pontrjagin classes of the bundle, while, for n
even, e(E) is called the Euler class of the bundle.
Remark 80. In particular, the three subsections apply for com-
plex, real, and oriented real vector bundles, where the terminology
Chern, Pontrjagin and Euler classes are usually applied (resp.),
50
by means of Section 35. By definition, the classes of a vector
bundle are the classes of its frame principal bundle.
Total Chern and Pontrjagin classes:
c(E) = 1 + c1(E) + c2(E) + · · · ∈ H∗(B),
p(E) = 1 + p1(E) + p2(E) + · · · ∈ H∗(B).
§37. The axiomatic approach
Suppose E1 ⊕ E2 is a Whitney sum of two (real or complex)
vector bundles. By the previous section, since the classes come
from determinants, the total class for E is the product of the
total classes of E1 and E2: c(E1 ⊕ E2) = c(E1) ∧ c(E2) (for
complex) p(E1 ⊕ E2) = p(E1) ∧ p(E2) (for real). Moreover,
extending the definition of the Euler class to odd dimensional
real vector bundles as 0, for the Euler class it also holds that
e(E1 ⊕ E2) = e(E1) ∧ e(E2); see [S], Vol.5, Ch.13, Theorem 22.
In particular: if the orientable vector bundle E has a nowhere
vanishing section, then e(E) = 0.
In fact, there is a way of defining characteristic classes for vector
bundles in an axiomatic way: we proved that they exist, and it
is not hard to see that they are unique. For example, for Chern
classes for complex vector bundles (CVB) we have:
• Axiom 1: For each CVB E over M , and each integer k ≥ 0,
there exist a class ck(E) ∈ H2k(M), with c0(E) = 1 (so we can
define c(E)=∑∞
i=0 ck(E)∈H∗(M), the total Chern class of E.)
• Axiom 2 (Naturality): IfE is a CVB overM and f : M ′ → M
is smooth, then c(f ∗E) = f ∗c(E).
51
• Axiom 3 (Whitney sum formula): If E,E ′ are CVBs over Mand E1⊕E ′ their Whitney sum, then c(E⊕E ′) = c(E)∧ c(E ′).• Axiom 4 (Normalization): If CP1 is the complex projective
line and P its canonical complex line bundle, then∫
CP1c1(P)=−1.
Exercise. Show that the Chern classes for CVBs as we defined satisfy
the four axioms above, thus proving existence; see §35 and [KN] II c.13.
Axiomatically, the Pontrjagin classes of a real vector bundle E
are defined simply by pk(E) = c2k(E ⊗ C).
For oriented real vector bundles of rank k, the Euler class is de-
fined with the same axioms as the Chern classes, except that, in
Axiom 1, we require e(E) ∈ Hk(M), and e(M) = 0 if k is odd.
§38. The Poincare-Hopf Theorem in all dimensions
Let X ∈ X(Mn) be a vector field on an oriented Mn with an
isolated singularity at p ∈ Mn. If we restrict X to the boundary
of a small ball Bǫ around p, we have Sn−1p := T1M ∩ TpM and
V = X/‖X‖ : ∂Bǫ∼= S
n−1 → T1Bǫ∼= Bǫ × S
n−1p
for some trivializing chart ϕ : Bǫ × Sn−1p → T1Bǫ with ϕ ip
being the inclusion Sn−1p ⊂ T1Bǫ, where ip : S
n−1p → Bǫ × S
n−1p ,
ip(v) = (p, v). We define the index of X at p as the integer
I(p) = deg(V ),
where V = π2 ϕ−1 V : ∂Bǫ∼= S
n−1 → Sn−1p . Notice that, for
ǫ small, V ∼= ϕ ip V as smooth functions.
52
With these definitions, the Poincare-Hopf Theorem 60 holds for
any compact oriented manifold and any vector field with isolated
singularities, and not just for surfaces. Indeed, by the proof of
the Gauss-Bonnet-Chern Theorem 83 below it follows that the
total index of a vector field is a topological invariant, i.e., does
not depend on the vector field. But it is easy to construct a
vector field whose total index is the Euler characteristic: for a
triangulation T , define VT as having precisely one singularity on
the ‘center’ of each simplex of T in a way so that the flow lines
of the vector field point from the centers of higher dimensional
simplexes towards the lower dimensional simplexes. Such a vector
field has total index equal to the Euler characteristic.
§39. The Gauss-Bonnet-Chern Theorem ([L])
Consider a compact oriented even dimensional Riemannian man-
ifold M 2m. Its tangent bundle is an SO(2m)-bundle, and so it
has its Euler Class, e(TM) ∈ H2m(M) ∼= R. So its integral∫
e(TM) ∈ R
is a topological invariant that does not depend on the Rie-
mannian metric. In terms of the curvature Ω of the Levi-Civita
connection, (12) ⇒ (2π)me(TM) is represented by the 2m-form
pf (Ω) =1
2mm!
∑
σ∈S2m
sign(σ) Ωσ1σ2∧ · · · ∧ Ωσ2m−1
σ2m.
Now, consider the sphere bundle π : T1M → M . Then,
π∗(pf (Ω)) ∈ Ω2m(T1M).
53
Lemma 81. (S.S.Chern; Transgression Lemma) There is
λ ∈ Ω2m−1(T1M) such that π∗(pf (Ω)) = dλ. In addition,∫
S2m−1p
λ|S2m−1p
= (2π)m, for all p ∈ M .
Proof: A long algebraic construction... See [L], Lemma 3.2.3.
Remark 82. For m = 1, since G = SO(2) ∼= S1 is abelian,
by (9), (10), (11) and Proposition 67 we can take λ = w in
Lemma 81.
Theorem 83. (Generalized Gauss-Bonnet-Chern Theorem)
If M 2m is compact and orientable, then∫
e(TM) = χ(M 2m).
Proof: Let X be a vector field with isolated singularities only
p1, . . . , pr. Remove small balls Bǫ(pi) from M , and define
Mǫ = M \ ∪iBǫ(pi) and V = X/‖X‖ : Mǫ → T1Mǫ. Then,
by the Transgression Lemma 81 and Stoke’s Theorem,∫
wpf = limǫ→0
∫
Mǫ
pf (Ω) = limǫ→0
∫
Mǫ
V ∗(π∗(pf (Ω)))
= limǫ→0
∫
Mǫ
V ∗(dλ) = limǫ→0
∫
Mǫ
d(V ∗λ)
=∑
i
limǫ→0
∫
∂Bǫ(pi)
V ∗λ =∑
i
limǫ→0
∫
∂Bǫ(pi)
V ∗((ϕ ipi)∗λ)
=∑
i
I(pi)
∫
S2m−1pi
λ|S2m−1pi
= (2π)m∑
i
I(pi).
Therefore, the total index is a topological invariant, independent
of the vector field X . But we saw in §38 that there exists a vectorfield whose total index is equal to χ(M 2m).
54
Remark 84. The above is essentially Chern’s original proof in
[C]. For an alternative proof using characteristic classes more
deeply, see [S], Vol 5, Ch. 13, Theorem 26.
Remark 85. Again, notice that we not only proved the Gauss-
Bonnet-Chern Theorem, but also Poincare-Hopf Theorem 60 for
any dimensions.
References
[C] S.S. Chern; A Simple Intrinsic Proof of the Gauss-
Bonnet Formula for Closed Riemannian Manifolds.
Ann. Math. 45 (4), 1944, 747-752.
[dC] M. do Carmo; Differential Geometry of Curves and Sur-
faces. Pearson Education Canada, 1976.
[H] A. Hubery; Notes on fibre bundles. Lecture notes here
with backup here.
[KN] S. Kobayashi, K. Nomizu; Foundations of Differential
Geometry. Wiley Classics Library, Volume I and II.
[L] Y. Li; The Gauss-Bonnet-Chern Theorem on Rieman-
nian Manifolds. ArXiV 1111.4972v4.
[MS] J. Milnor, J. Stasheff; Characteristic classes. Princeton
University Press, 1974.
[S] M. Spivak; A comprehensive introduction to differential
geometry. Publish or Perish Inc., Texas, 1975.
55